This fixes #20613 and is compatible with existing behavior of platforms, device types, and device indices. It fixes the case of multiple platforms when those platforms have common substrings. See issue for details.

There should be no change in already working existing behavior of 3.4 and 4.x master when using APIs that allow specification of OpenCL "configuration" or setting the envvar OPENCV_OPENCL_DEVICE. If it worked before, it should continue working in the same way. It it previously failed and could only use the workaround I describe in the issue...now the PR's code will work using the expected manner + the workaround.

Code approach:

• copy 4.x selectOpenCLDevice() back to 3.4 -- it is compatible and easily used
• filter platforms rather than trying only first matching platform
• Alter code to interate the platforms when attempting to get specific devices
• Overall, there is very little code change.
• This PR works on 3.4 and 4.x master
• I squashed the two commits (4.x code, the fix) into one

I could not think of a consistent way to create a test case to exercise this. So existing test cases and CI processes should continue to work. It should not cause a failure.

It is possible to create a specific test case for a specific machine. So if you want that, I will need to have a specific CI setup for the specific fail case described in the issue. Then I can write a test case for that scenario to catch regressions on that specific CI setup.

Pull Request Readiness Checklist

• I agree to contribute to the project under Apache 2 License.
• To the best of my knowledge, the proposed patch is not based on a code under GPL or other license that is incompatible with OpenCV
• The PR is proposed to proper branch
• There is reference to original bug report and related work
• There is accuracy test, performance test and test data in opencv_extra repository, if applicable
  Patch to opencv_extra has the same branch name.
• The feature is well documented and sample code can be built with the project CMake